| { | |
| "_name_or_path": "nvidia/Nemotron-Mini-4B-Instruct", | |
| "architectures": [ | |
| "NemotronForCausalLM" | |
| ], | |
| "attention_bias": false, | |
| "attention_dropout": 0.0, | |
| "bos_token_id": 2, | |
| "compression_config": { | |
| "config_groups": { | |
| "group_0": { | |
| "input_activations": { | |
| "block_structure": null, | |
| "dynamic": true, | |
| "group_size": null, | |
| "num_bits": 8, | |
| "observer": "memoryless", | |
| "observer_kwargs": {}, | |
| "strategy": "token", | |
| "symmetric": true, | |
| "type": "float" | |
| }, | |
| "output_activations": null, | |
| "targets": [ | |
| "Linear" | |
| ], | |
| "weights": { | |
| "block_structure": null, | |
| "dynamic": false, | |
| "group_size": null, | |
| "num_bits": 8, | |
| "observer": "minmax", | |
| "observer_kwargs": {}, | |
| "strategy": "channel", | |
| "symmetric": true, | |
| "type": "float" | |
| } | |
| } | |
| }, | |
| "format": "float-quantized", | |
| "global_compression_ratio": 1.2473985521511244, | |
| "ignore": [ | |
| "lm_head" | |
| ], | |
| "kv_cache_scheme": null, | |
| "quant_method": "compressed-tensors", | |
| "quantization_status": "frozen" | |
| }, | |
| "eos_token_id": 3, | |
| "head_dim": 128, | |
| "hidden_act": "relu2", | |
| "hidden_size": 3072, | |
| "initializer_range": 0.0134, | |
| "intermediate_size": 9216, | |
| "kv_channels": 128, | |
| "max_position_embeddings": 4096, | |
| "mlp_bias": false, | |
| "model_type": "nemotron", | |
| "norm_eps": 1e-05, | |
| "num_attention_heads": 24, | |
| "num_hidden_layers": 32, | |
| "num_key_value_heads": 8, | |
| "partial_rotary_factor": 0.5, | |
| "rope_theta": 10000, | |
| "tie_word_embeddings": false, | |
| "torch_dtype": "bfloat16", | |
| "transformers_version": "4.44.1", | |
| "use_cache": true, | |
| "vocab_size": 256000 | |
| } | |